causally.scm.context.SCMContext.autoregressive_model

SCMContext.autoregressive_model(order: int)

Make the assumption of model with time lagged autoregressive effects.

Structural equations take the following autoregressive form:

\[X_i(t):= f_i(\operatorname{PA}_i(t)) + N_i + \sum_{k=t-\operatorname{order}} \alpha(k) X_i(k)\]

where \(f_i\) is the nonlinear causal mechanism, \(N_i\) is the noise term of the structural equation, \(\alpha(k)\) is a coefficient uniformly sampled between -1 and 1, \(t\) the index of a sample, interpreted as the time step.

Parameters:

order (int) – The number of time lags

Notes

Time lagged autoregressive effects are not supported for the post-nonlinear model.